ECJ-to-Java converter

ECJ will generate Lisp output, by default, but wouldn't it be nice if it could generate pure Java code for you — automatically? ECJ2Java allows you to do exactly that, with a minimum amount of disruption to your existing GP functions. It fits into your existing ECJ framework easily too.

What to do

All you need to do is extend either ParseableGPNode, or ParseableERC instead of ECJ's default GPNode and ERC classes, and implement the required getJavaCode() function. You can then use the ECJ2Java program to automatically convert a GPIndividual tree into a Java class. You can do this in the describe method of your Problem class's code.

A full, standalone, example is provided in ac.essex.ooechs.ecj2java.example. The example shows a equation solving program that outputs its individuals as Java files. Have a look at its solution to the Pythagoras Equation.

What to do in detail

These are the essential steps:

Modify your function and terminal classes such that they extend ParseableGPNode instead of GPNode.
Modify your ERC classes to extend ParseableERC instead of ERC.
Implement the getJavaCode() function in each. Have a look at some examples: Add.java, Divide.java.
Create an extended KozaStatistics class that calls the describe method on your problem. Ensure the KozaStatistics class is referred to in the params file.
Create a describe method in your Problem class (if not already present).
Add code similar to that in listing 1 to the describe method (below).

Listing 1 — Example describe method

static int counter = 0;

public void describe(final Individual ind, final EvolutionState state,
  final int threadnum, final int log, final int verbosity)
{
    state.output.println("

Best Individual", verbosity, log);

    KozaFitness f = ((KozaFitness) ind.fitness);

    counter++;

    /**
     * Create a ECJ2Java Writer here, with the following parameters:
     */
    String classPackage = "ac.essex.ooechs.ecj2java.example.individuals";
    String className = "MathsSolution" + counter;
    String functionSignature = "public double calculate(double x)";
    String comment = "Fitness: " + f.hits;
    JavaWriter writer = new JavaWriter(className, functionSignature,
      comment, classPackage);

    /**
     * Decide where the Java file is to be saved:
     */
    File saveTo = new File("/home/ooechs/example/individuals/");

    /**
     * Generate and save the file
     */
    try {
        writer.saveJavaCode((GPIndividual) ind, saveTo);
    } catch (IOException e) {
        e.printStackTrace();
    }

    /**
     * And that's all there is to it!
     */
    System.out.println("TP: " + f.hits);
    System.out.println("Fitness: " + f.standardizedFitness());
}

Have a look at some code generated automatically: MathSolution1.java.

Downloads

Please note that ParseableERC.java refers to a class in the author's commons library that is currently unavailable. You'll need to change the import statement to import your own Data class. There is a compatible one here.

The example may be run with the following command:
java ec.Evolve -file [full/path/to]example.params -p stat.gather-full=true